Bird and Drone Image Classification Using ResNet CNN: A Deep Learning Approach for Aerial Surveillance

Abstract

Accurate classification of bird and drone images is crucial in supporting aerial surveillance and security systems, particularly to distinguish between natural objects such as birds and man-made objects such as drones. Manual classification methods have limitations in terms of speed and accuracy, thus necessitating a more efficient and reliable technology-based approach. This study aims to implement a ResNet-50 based Convolutional Neural Network (CNN) architecture to automatically classify bird and drone images. The dataset used was obtained from the Kaggle platform and consists of two classes: Bird and Drone, with a total of 22,407 images. The data was split into training (17,323 images), testing (844 images), and validation (1,740 images). All images underwent preprocessing and augmentation steps to enhance data quality and model training performance. The model was developed using the ResNet-50 architecture, which is well-regarded for handling complex image classification tasks. Evaluation results show that the model achieved an accuracy of 92%. For the Bird class, a precision of 0.83 and a recall of 0.99 were obtained, while for the Drone class, precision reached 0.99 and recall was 0.86. The average F1-score of 0.92 indicates that the model delivers balanced and reliable performance in the binary image classification task.